Here's a collection of highlights, selected totally subjectively, from this week's HPC news stream as reported at insideHPC.com and HPCwire.
>>10 words and a link
Sun announces two- and four-socket Xeon blades at IDF,
15 European countries announce “pan-European” HPC pact,
DOE announces computational science publication and attempts to revive the word “webzine,”
IBM promises two POWER systems for Dutch National Supercomputing Center,
Mellanox previews 40 Gb/s InfiniBand adapter,
>>Intel renames Gesher
Intel's Developer Forum was this week in Beijing, and so there was a flood of news in the technical press around the chipmaker and its upcoming technology. (HPCwire included a lot of the important announcements at http://www.hpcwire.com/hpc/1436533.html.)
Of note for those of you playing along in the home version of the Intel chip family name game is that Intel has abandoned the name “Gesher” for the architecture following Nehalem. Henceforth it shall be known as “Sandy Bridge.” Intel's chief Pat Gelsinger said in an interview with The Register that they abandoned Gesher “for a variety of internal and external reasons” and added that the company prefers “non-volatile code-names.”
Gesher means “bridge” in Hebrew and is also the name of an Israeli political party.
>>Intel's reconfigurable hardware
In addition to its own product development, Intel has to make its metal more accessible to third party developers. Intel announced this week that it has added XtremeData to its team of FPGA partners. Xilinx and Altera were already in the stable and announced products that put accelerators near the chip on Intel boards.
This is reminiscent of what AMD is doing with its Torrenza initiative, but looks to me to be less open than AMD's effort. More at http://www.theregister.co.uk/2007/04/17/intel_idf_serverstuff/.
>>Patching hardware over the internet
Josep Torrellas of the University of Illinois at Urbana Champaign is working on a new system for correcting defective computer hardware without cracking open the case and heating up a soldering iron.
The system, called Phoenix, works by adding special hardware to a chip that can be reprogrammed to correct hardware faults. When a fault is discovered in a Phoenix-enabled chip, the manufacturer would release a patch containing the defect signature to all computers affected with the flaw. The Phoenix system then monitors the chip, watching for the operation signature that exercises the defective portion of the hardware and performs new, correct operations instead. Existing chips Crusoe (from Transmeta) and the Itanium can also be patched, but Torrellas claims his approach is more effective.
Of course some have raised the inevitable question: if hardware vendors can easily fix after they ship, will this induce them to spend less time trying to get it right in the first place? Full article at http://www.technologyreview.com/Infotech/18513/.
John West summarizes the headlines in HPC every day at insideHPC.com, and writes on leadership and career issues for technology professionals at InfoWorld and on his own blog at onlytraitofaleader.com. You can contact him at [email protected].